From P Dickam instruction sheet https://pauldickman.com/talk/proport...22/melanoma.do
On here:
From GOF, Schoenfelds - gender & stage have a value of 0 and schoenfelds are not a straight line.
P Dickman, continue to relax PH assumption for gender ... to compare the results from STCOX with tvc(male) vs STCOX -no tvc - vs STPM2 tvc(male)
He shows us one gets similar results.
However why do this, when
1. 'Stage' - schoenfeld results are much worse, GOF = 0.00
2. I tried the stcox tvc(stage) --> obtained a p value of >0.05 - showing its not a timevarying covariate
My question
1. Why try show the results are comparable using gender - which the PH assumption isn't particularly violated as shown in schoenfelds + gof
2. If tvc(stage) - p value is >0.05; showing it's not a timevarying covariate. However when one plots GOF, schonfelds , the results are perfect - horizontal lines and GOF >0.00
Unlike when not introducing stage as a tvc
3. However, the results of introducing stage as a tvc; gender as a tvc; gender without tvc, stage without a tvc. The HR are all similar.
So my question, does it really matter trying to perfect out model with PH assumptions.... is this similar to our obsession of p values?
On here:
From GOF, Schoenfelds - gender & stage have a value of 0 and schoenfelds are not a straight line.
P Dickman, continue to relax PH assumption for gender ... to compare the results from STCOX with tvc(male) vs STCOX -no tvc - vs STPM2 tvc(male)
He shows us one gets similar results.
However why do this, when
1. 'Stage' - schoenfeld results are much worse, GOF = 0.00
2. I tried the stcox tvc(stage) --> obtained a p value of >0.05 - showing its not a timevarying covariate
My question
1. Why try show the results are comparable using gender - which the PH assumption isn't particularly violated as shown in schoenfelds + gof
2. If tvc(stage) - p value is >0.05; showing it's not a timevarying covariate. However when one plots GOF, schonfelds , the results are perfect - horizontal lines and GOF >0.00
Unlike when not introducing stage as a tvc
3. However, the results of introducing stage as a tvc; gender as a tvc; gender without tvc, stage without a tvc. The HR are all similar.
So my question, does it really matter trying to perfect out model with PH assumptions.... is this similar to our obsession of p values?
Attached Files
